Real-Time Merchant Recommendation System
Prototype
Illustrative re-ranking over mock merchants — the production model used XGBoost classifiers and stacking meta-learners on real transactional and geolocation features.
Highlights
- Built a real-time predictive recommendation model designed to pair active credit card clients with prospective local merchants based on transactional history and geolocation criteria.
- Engineered cross-functional feature pipelines by performing rigorous correlation filters, missing-value interpolations, and feature distributions.
- Trained optimized XGBoost classifiers and stacking meta-learners, generating a 5% leap in absolute predictive AUC accuracy and boosting target activation metrics by ~11% among top 10 recommended nodes.
Technologies
- XGBoost
- Stacking Meta-Learners
- Correlation Filters
- Feature Distributions